Enabling Morally Sensitive Robotic Clarification Requests

The design of current natural language-oriented robot architectures enables certain architectural components to circumvent moral reasoning capabilities. One example this is reflexive generation clarification requests as soon referential ambiguity detected in a human utterance. As shown previous research, can lead robots (1) miscommunicate their dispositions and (2) weaken perception or application norms within context. We present solution these problems by performing on each potential disambiguation an ambiguous utterance responding accordingly, rather than immediately naively requesting clarification. implement our the Distributed Integrated Cognition Affect Reflection architecture, which, knowledge, only architecture with both request then evaluate method subjects experiment, results which indicate that approach successfully ameliorates two identified concerns.